In Part 1, I shared how a dusty screenplay led me to dive into vibe coding and build an AI-powered writers room. In Part 2, I walked through what breaks when instructing multiple AI personas to collaborate, and why constraint matters more than capability.
What I Learned
When I built my AI writers room, I thought the personas were going to hand me ideas. Funnier dialogue. Stronger character backstories. Smarter story arcs. Richer worlds. That's not what happened.
The value wasn't what the AI generated - it was what the AI catalyzed. The process of engaging with multiple perspectives, being asked the right question at the right moment, having my assumptions challenged - that's what unlocked better work.
When I uploaded that old pilot script (see Part 1) into the app, I started down a collaborative writing path. At one point I was stuck on a new scene. Something wasn't landing. The scene felt flat.
One of the personas asked a simple question: “What was this character’s state before? What's the wound he’s carrying into every moment of his life?”
It didn't give me the answer. It challenged me to go deeper into the character’s past, when I was trying to solve a story problem to move forward.
And suddenly I saw it - the past conflict, the suffering that explained everything about his present. I wrote a scene in a matter of minutes that made his motivation clear. No character explaining himself. Just behavior that made sense because now I understood where he came from. I wrote that, with an AI spark.
The process of being asked the right question at the right moment unlocked something I couldn't access alone. I generated more original ideas from engaging with the AI than were ever handed to me by the personas. The AI didn't write my screenplay. The AI changed how I worked on my screenplay.
The Outputs Trap
Right now, people and companies are asking the same question: "How should AI be used?" And when asking this question, the focus tends to be on outputs: "Look what it created!" "I knocked out this brief in 30 seconds!" "Check out this image/video/code!" This makes sense. Outputs are visible. They're measurable. They're demo-able. They can substantiate a KPI. And many are truly extraordinary. When you're trying to justify an AI investment, you show what it produced, and the value it created.
Compelling and important, for sure. But in my experience working with AI, I’ve learned that it's not about the outputs - it’s about the process.
If you're leading AI adoption at your company, ask different questions: "How can AI change how we work?" vs. "What can AI produce for us?" Instead of measuring outputs generated, measure insights unlocked, assumptions challenged, new approaches discovered.
Start small. Pick one process where you're stuck - strategy sessions, creative reviews, data analysis, problem-solving. Add AI not as a content generator, but as a thinking partner. Give it constraints, assign it a role, make it push back on your ideas.
The outputs are easy to measure. The process is where value hides.
The Orchestration Opportunity
I’m discovering that agents and orchestration is where the magic happens.
It’s not just about working with the best models. Models are commoditizing fast. What Claude does today, Gemini will do tomorrow. Even with the innovation race that is underway, the capability gap is closing.
Those who enable agents AND orchestration, and coordinate AI capabilities in processes that are functional, relevant, and improve how people work, are going to ride the innovation wave. Utilizing the “right model, in the right place, at the right time.”
That means:
- Multi-model routing: Utilizing various AI models and agents in the enterprise tech stack, employed for different tasks based on their strengths
- Productive constraint: Teaching AI when to contribute and when to stay silent
- Human-in-the-loop design: AI proposes, human decides, human+AI generates. Not AI generates, human accepts.
- Process integration: AI embedded in workflows, not bolted on as a separate tool
It requires understanding how work happens (operations, systems, processes), the roles people play in that work, and where AI can elevate it. Not just speed it up.
A Practitioner's Perspective
The hard part wasn't making AI capable. It was making AI useful, which often meant making it less capable in targeted ways. Less eager to please. Less likely to volunteer ideas. More willing to be constrained… to stay silent.
The value wasn't in what AI produced. It was in how AI changed my creative process, taught me concepts, and sparked ideas. I got better as a writer.
And the real challenge wasn't technical. It was designing for collaboration, not generation (frankly, I’m still tuning this.)
If you're evaluating AI tools, I'd encourage you to ask:
- Does this help us produce more stuff, or does it help us think better?
- Is this designed for generation or collaboration? What do we need in this situation?
- Will this support how our people work and be adopted, or be another tool to the stack?
From Dusty Screenplay to Business Strategy
Three months ago, I found an old pilot script in a drawer and wondered if AI could help me improve it. I thought I'd get a better script handed to me. Instead, I got something more valuable: a new way of working.
That dusty screenplay became a writers’ room, which became an app. The app became a lesson in what AI is capable of - not replacing human creativity, but catalyzing it.
The same principle that made my fictional characters more compelling makes real businesses more innovative: it's not about what AI produces for you. It's about how AI changes what you're capable of producing.
Your next breakthrough won't come from what AI writes, codes, or creates. It'll come from the question AI helps you ask that you never thought to ask before. That's the real competitive advantage. And it starts with changing one simple question: not "what can AI do for me?" but "how can AI help me think?"
The rest writes itself.